TABX: A High-Throughput Sandbox Battle Simulator for Multi-Agent Reinforcement Learning

TL;DR

TABX introduces a high-throughput, JAX-based sandbox for multi-agent reinforcement learning that emphasizes configurability and scalability. By exposing units, terrain zones, heuristic policies, and physics as parameterized controls, it enables systematic exploration of information-dependent value learning, long-horizon exploration, and zero-shot generalization, supported by a GUI scenario editor and vectorized GPU execution. The paper demonstrates a suite of MARL and unsupervised environment design baselines, analyzes centralized value learning versus decentralized approaches, and shows TABX achieves substantial throughput gains over prior frameworks while facilitating rigorous experimental control. This framework offers a practical, reproducible platform for probing how environment design interacts with MARL algorithms, with potential to accelerate research in robust coordination, generalization, and efficient training. The work lays groundwork for future expansions such as LOS-restricting terrain, fortifications, pixel observations, and richer unitSkill dynamics.

Abstract

The design of environments plays a critical role in shaping the development and evaluation of cooperative multi-agent reinforcement learning (MARL) algorithms. While existing benchmarks highlight critical challenges, they often lack the modularity required to design custom evaluation scenarios. We introduce the Totally Accelerated Battle Simulator in JAX (TABX), a high-throughput sandbox designed for reconfigurable multi-agent tasks. TABX provides granular control over environmental parameters, permitting a systematic investigation into emergent agent behaviors and algorithmic trade-offs across a diverse spectrum of task complexities. Leveraging JAX for hardware-accelerated execution on GPUs, TABX enables massive parallelization and significantly reduces computational overhead. By providing a fast, extensible, and easily customized framework, TABX facilitates the study of MARL agents in complex structured domains and serves as a scalable foundation for future research. Our code is available at: https://anonymous.4open.science/r/TABX-00CA.

Figures (34)

• Figure 1: An illustrative scenario showcasing core features of TABX: (a) fan-shaped partial observability, (b) non-targeted interactions, (c) heterogeneous unit roles, and (d) terrain zones that impose complex strategic demands.
• Figure 2: Overview of the TABX scenario editor. The interface enables visual authoring of scenarios by allowing users to place ally and enemy units, configure unit specifications, and define environmental zones with adjustable functional effects. The editor provides direct access to key environment parameters through an interactive, code-free workflow.
• Figure 3: Representative designed scenarios illustrating different degrees of dependence on global state information. Colored dashed lines indicate the fan-shaped fields of view of individual allies, highlighting how partial observability and viewpoint separation vary across scenarios. Colored ellipses represent terrain zones with distinct functional effects, such as movement speed reduction or visibility occlusions. Allies are denoted by a red outline, while enemies are indicated by a green outline.
• Figure 4: Average episode win rates for baseline algorithms across eight different scenarios.
• Figure 5: Value estimation error $\lvert V - V^* \rvert$ across different scenarios. $V^*$ is approximated via extensive rollout simulations. Results compare centralized value learning (MAPPO) with independent learning (IPPO).